Author: Robert Hyatt
Date: 10:10:23 09/03/03
Go up one level in this thread
On September 03, 2003 at 09:06:39, Gian-Carlo Pascutto wrote: >On September 03, 2003 at 09:02:28, Matthew Hull wrote: > >>But for chess, each piece of the search process will be on it's own processor >>and all the data it needs to do the search will be copied once at the start of >>it's "assignment", then it's off to the races. The only time he needs to use >>non-local memory is at the start (copy) and the end (report results). >> >>Yes? > >Nope :) > >You need to constantly check for abort-fail-high and abort-adjust-alpha >conditions on the other processors. I think most implementations will need >to constantly check/update what moves are left to process as well. > >-- >GCP The first is easy to solve. I have an "abort" flag, but it is in the TREE structure. That means it is _always_ a local memory reference for the processor and there is no overhead. Also, you don't need to "constantly check/update what moves are left". The root position at a split point is a tiny part of the work each processor is going to do.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.